chore: validate when quantization is 'none'

Author: himanshusinghs

src/common/search/vectorSearchEmbeddingsManager.ts

@@ -198,9 +198,33 @@ export class VectorSearchEmbeddingsManager {
         }
 
         switch (definition.quantization) {
-            // Because quantization is not defined by the user
-            // we have to trust them in the format they use.
+            // Quantization "none" means no quantization is performed, so
+            // full-fidelity vectors are stored therefore the underlying vector
+            // must be stored as an array of numbers having the same dimension
+            // as that of the index.
             case "none":
+                if (!Array.isArray(fieldRef)) {
+                    return constructError({
+                        error: "not-a-vector",
+                    });
+                }
+
+                if (fieldRef.length !== definition.numDimensions) {
+                    return constructError({
+                        actualNumDimensions: fieldRef.length,
+                        actualQuantization: "none",
+                        error: "dimension-mismatch",
+                    });
+                }
+
+                if (!fieldRef.every((e) => this.isANumber(e))) {
+                    return constructError({
+                        actualNumDimensions: fieldRef.length,
+                        actualQuantization: "none",
+                        error: "not-numeric",
+                    });
+                }
+
                 return undefined;
             case "scalar":
             case "binary":

tests/unit/common/search/vectorSearchEmbeddingsManager.test.ts

@@ -42,6 +42,13 @@ const embeddingConfig: Map<EmbeddingNamespace, VectorFieldIndexDefinition[]> = n
     [
         mapKey,
         [
+            {
+                type: "vector",
+                path: "embedding_field_wo_quantization",
+                numDimensions: 8,
+                quantization: "none",
+                similarity: "euclidean",
+            },
             {
                 type: "vector",
                 path: "embedding_field",
@@ -278,51 +285,74 @@ describe("VectorSearchEmbeddingsManager", () => {
                     expect(result).toHaveLength(0);
                 });
 
-                it("documents inserting the field with wrong type are invalid", async () => {
-                    const result = await embeddings.findFieldsWithWrongEmbeddings(
-                        { database, collection },
-                        { embedding_field: "some text" }
-                    );
-
-                    expect(result).toHaveLength(1);
-                });
+                it.each(["embedding_field", "embedding_field_wo_quantization"] as const)(
+                    "documents inserting the field with wrong type are invalid - $0",
+                    async (field) => {
+                        const result = await embeddings.findFieldsWithWrongEmbeddings(
+                            { database, collection },
+                            { [field]: "some text" }
+                        );
 
-                it("documents inserting the field with wrong dimensions are invalid", async () => {
-                    const result = await embeddings.findFieldsWithWrongEmbeddings(
-                        { database, collection },
-                        { embedding_field: [1, 2, 3] }
-                    );
-
-                    expect(result).toHaveLength(1);
-                    const expectedError: VectorFieldValidationError = {
-                        actualNumDimensions: 3,
-                        actualQuantization: "scalar",
-                        error: "dimension-mismatch",
-                        expectedNumDimensions: 8,
-                        expectedQuantization: "scalar",
-                        path: "embedding_field",
-                    };
-                    expect(result[0]).toEqual(expectedError);
-                });
+                        expect(result).toHaveLength(1);
+                    }
+                );
 
-                it("documents inserting the field with correct dimensions, but wrong type are invalid", async () => {
-                    const result = await embeddings.findFieldsWithWrongEmbeddings(
-                        { database, collection },
-                        { embedding_field: ["1", "2", "3", "4", "5", "6", "7", "8"] }
-                    );
+                it.each([
+                    { path: "embedding_field", expectedQuantization: "scalar", actualQuantization: "scalar" },
+                    {
+                        path: "embedding_field_wo_quantization",
+                        expectedQuantization: "none",
+                        actualQuantization: "none",
+                    },
+                ] as const)(
+                    "documents inserting the field with wrong dimensions are invalid - path = $path",
+                    async ({ path, expectedQuantization, actualQuantization }) => {
+                        const result = await embeddings.findFieldsWithWrongEmbeddings(
+                            { database, collection },
+                            { [path]: [1, 2, 3] }
+                        );
+
+                        expect(result).toHaveLength(1);
+                        const expectedError: VectorFieldValidationError = {
+                            actualNumDimensions: 3,
+                            actualQuantization,
+                            error: "dimension-mismatch",
+                            expectedNumDimensions: 8,
+                            expectedQuantization,
+                            path,
+                        };
+                        expect(result[0]).toEqual(expectedError);
+                    }
+                );
 
-                    expect(result).toHaveLength(1);
-                    const expectedError: VectorFieldValidationError = {
-                        actualNumDimensions: 8,
-                        actualQuantization: "scalar",
-                        error: "not-numeric",
-                        expectedNumDimensions: 8,
-                        expectedQuantization: "scalar",
-                        path: "embedding_field",
-                    };
-
-                    expect(result[0]).toEqual(expectedError);
-                });
+                it.each([
+                    { path: "embedding_field", expectedQuantization: "scalar", actualQuantization: "scalar" },
+                    {
+                        path: "embedding_field_wo_quantization",
+                        expectedQuantization: "none",
+                        actualQuantization: "none",
+                    },
+                ] as const)(
+                    "documents inserting the field with correct dimensions, but wrong type are invalid - $path",
+                    async ({ path, expectedQuantization, actualQuantization }) => {
+                        const result = await embeddings.findFieldsWithWrongEmbeddings(
+                            { database, collection },
+                            { [path]: ["1", "2", "3", "4", "5", "6", "7", "8"] }
+                        );
+
+                        expect(result).toHaveLength(1);
+                        const expectedError: VectorFieldValidationError = {
+                            actualNumDimensions: 8,
+                            actualQuantization,
+                            error: "not-numeric",
+                            expectedNumDimensions: 8,
+                            expectedQuantization,
+                            path,
+                        };
+
+                        expect(result[0]).toEqual(expectedError);
+                    }
+                );
 
                 it("documents inserting the field with correct dimensions and quantization in binary are valid", async () => {
                     const result = await embeddings.findFieldsWithWrongEmbeddings(